Using AI in products
Somewhere along the way, we equated “AI” with “conversation.” Chat interfaces became the face of intelligence. Every demo, every product launch, every new app is all chat.
It's an easy metaphor. Talking feels human. But it's also the most primitive way to interact with intelligence.
Real AI doesn't ask for your attention. It just works. Quietly, contextually, invisibly.
1. The Problem: Why Chat Isn't Enough
Chat is linear. It's a turn-based interface pretending to be fluid. You ask, it answers. You correct, it rephrases. You guide, it follows. It's a loop.
But software isn't a conversation. It's a system of intentions. Users don't want to talk about their goals. They want to achieve them.
2. The Shift: From Conversation to Context
The best products anticipate. They see what you're trying to do and act before you ask.
In Working with AI, I wrote that AI should work like a teammate who understands context. One who can fill in the blanks, handle the boring parts, and know when to step in.
That same principle applies to product design. AI shouldn't wait for explicit commands. It should infer, assist, and adapt.
Google has done this quietly for years. Translate text. Autocomplete your thoughts. Detect spam. Summarize what matters. These aren't “AI features.” They're product instincts.
The best AI understands what matters and adapts to it.
3. The Pattern Library: Designing AI Behaviors
The next generation of apps won't bolt AI on top. They'll weave it through every layer.
Here's what that looks like in practice:
- Suggestions - predict intent before input (Linear's AI issue labels)
- Summaries - compress complexity into clarity (Google AI Overview)
- Actions - execute what's implied, not just what's said
- Adaptation - tune UI and data to match the user's moment
- Memory - build long-term understanding without asking for it
4. The Craft: Making AI Feel Invisible
The goal isn't to make AI visible. It's to make software feel effortless.
Every AI touchpoint should feel like design.
That means setting latency budgets, defining confidence thresholds, and designing what happens when the model fails.
When we design this way, AI stops being a spectacle and starts being infrastructure. It becomes part of the product's rhythm. As natural as animation, as invisible as caching.
5. The Hard Part: Building Products with AI
Building AI products well is hard. Incredibly hard.
You can't just drop a model in and call it done. You're designing for uncertainty, context, and interpretation. It's not about accuracy. It's about fit.
Traditional product design is deterministic. You design states, flows, outcomes. AI design is probabilistic. You're orchestrating behavior. You're shaping a system that learns, not a feature that executes.
This shift changes everything. How you test, how you measure, how you think. You're no longer designing just for users, but with a model in the loop.
Hard isn't impossible, though. The same discipline that makes a product feel fast, clear, and delightful applies here too. Design for edge cases, build tight feedback loops, and treat the model like a collaborator that needs guidance, not freedom.
If done right, your product doesn't just “use AI.”
6. The Future: Every Product Becomes Adaptive
This is where things are heading. Products that feel alive. Interfaces that shift with you. Workflows that understand what you meant, not just what you clicked.
We're not designing prompts anymore.